Searcharxiv⌕ Search

arXiv subjects

Stephen Jewson

Publications and source records attributed to Stephen Jewson.

52 records · Page 3Linked to original sources
arXiv2004

Probabilistic forecasting of temperature: comments on the Bayesian Model Averaging approach

A specific implementation of Bayesian model averaging has recently been suggested as a method for the calibration of ensemble temperature forecasts. We point out the similarities between this new approach and an earlier method known as kernel regression. We also argue that the Bayesian model averaging method (as applied) has a number of flaws that would result in forecasts with suboptimally calibrated mean and uncertainty.

physics.ao-ph↗
arXiv2004

Singular vector ensemble forecasting systems and the prediction of flow dependent uncertainty

The ECMWF ensemble weather forecasts are generated by perturbing the initial conditions of the forecast using a subset of the singular vectors of the linearised propagator. Previous results show that when creating probabilistic forecasts from this ensemble better forecasts are obtained if the mean of the spread and the variability of the spread are calibrated separately. We show results from a simple linear model that suggest that this may be a generic property for all singular vector based ensemble forecasting systems based on only a subset of the full set of singular vectors.

physics.ao-ph↗
arXiv2004

The problem with the Brier score

The Brier score is frequently used by meteorologists to measure the skill of binary probabilistic forecasts. We show, however, that in simple idealised cases it gives counterintuitive results. We advocate the use of an alternative measure that has a more compelling intuitive justification.

physics.ao-ph↗
arXiv2004

Use of the likelihood for measuring the skill of probabilistic forecasts

We define the likelihood and give a number of justifications for its use as a skill measure for probabilistic forecasts. We describe a number of different scores based on the likelihood, and briefly investigate the relationships between the likelihood, the mean square error and the ignorance.

physics.ao-ph↗
arXiv2003

Comparing the ensemble mean and the ensemble standard deviation as inputs for probabilistic medium-range temperature forecasts

We ask the following question: what are the relative contributions of the ensemble mean and the ensemble standard deviation to the skill of a site-specific probabilistic temperature forecast? Is it the case that most of the benefit of using an ensemble forecast to predict temperatures comes from the ensemble mean, or from the ensemble spread, or is the benefit derived equally from the two? The answer is that one of the two is much more useful than the other.

physics.ao-ph↗
arXiv2003

Do probabilistic medium-range temperature forecasts need to allow for non-normality?

The gaussian spread regression model for the calibration of site specific ensemble temperature forecasts depends on the apparently restrictive assumption that the uncertainty around temperature forecasts is normally distributed. We generalise the model using the kernel density to allow for much more flexible distribution shapes. However, we do not find any meaningful improvement in the resulting probabilistic forecast when evaluated using likelihood based scores. We conclude that the distribution of uncertainty is either very close to normal, or if it is not close to normal, then the non-normality is not being predicted by the ensemble forecast that we test.

physics.ao-ph↗
arXiv2003

Five guidelines for the evaluation of site-specific medium range probabilistic temperature forecasts

Probabilistic temperature forecasts are potentially useful to the energy and weather derivatives industries. However, at present, they are little used. There are a number of reasons for this, but we believe this is in part due to inadequacies in the methodologies that have been used to evaluate such forecasts, leading to uncertainty as to whether the forecasts are really useful or not and making it hard to work out which forecasts are best. To remedy this situation we describe a set of guidelines that we recommend should be followed when evaluating the skill of site-specific probabilistic medium range temperature forecasts. If these guidelines are followed then the results of validation can be used directly by forecast users to make decisions about which forecasts to use. If they are not followed then the results of validation may be interesting, but will not be practically useful for users. We find that none of the published studies that evaluate such forecasts fall within our guidelines, and that, as a result, none convey the information that the users need to make appropriate decisions about which forecasts are best.

physics.ao-ph↗
arXiv2003

Moment based methods for ensemble assessment and calibration

We describe various moment-based ensemble interpretation models for the construction of probabilistic temperature forecasts from ensembles. We apply the methods to one year of medium range ensemble forecasts and perform in and out of sample testing. Our main conclusion is that probabilistic forecasts derived from the ensemble mean using regression are just as good as those based on the ensemble mean and the ensemble spread using a more complex calibration algorithm. The explanation for this seems to be that the predictable component of the variability of the forecast uncertainty is only a small fraction of the total forecast uncertainty. Users of ensemble temperature forecasts are advised, until further evidence becomes available, to ignore the ensemble spread and build probabilistic forecasts based on the ensemble mean alone.

physics.ao-ph↗
arXiv2003

The Assessment and Calibration of Ensemble Seasonal Forecasts of Equatorial Pacific Ocean Temperature and the Predictability of Uncertainty

We evaluate the performance of two 44 year ensemble seasonal hindcast time series for the Nino3 index produced as part of the DEMETER project. We show that the ensemble mean carries useful information out to six months. The ensemble spread, however, only carries useful information out to four months in one of the models, and two months in the other.

physics.ao-ph↗